In focal animal samples of the social interactions of individuals in groups, the possible variability of behaviour in time poses a problem of proper data selection. Four models are developed to reveal what conditions the behaviour under study must satisfy in order for data from the focal samples of both members of an interacting dyad to be used in estimating one member's partner distribution. The conditions are: the probability with which the subject animal acts on each partner, given that he acts at all, remains constant in time and the subject's overall activity rate remains constant or varies only within certain prescribed limits. If these conditions are not met, an animal's partner distribution should be estimated only from data recorded during his focal samples. The paper provides a means of testing whether a particular set of data conforms to these conditions.